License plate image enhancement method and device, equipment and storage medium

Abstract

The invention provides a license plate image enhancement method and device, equipment and a storage medium. The method comprises the following steps: acquiring a first position relationship among vertexes of a license plate to be enhanced and a second position relationship among vertexes of a license plate position in a to-be-fused background image; calculating a perspective transformation matrix between the to-be-enhanced license plate and the to-be-fused background image according to the first position relation and the second position relation; and fusing the to-be-enhanced license plate to the license plate position of the to-be-fused background image according to the transformation matrix to obtain an enhanced license plate image. According to the method, the perspective distortion is effectively reduced in a mode of fusing the license plate and the real background image through the perspective transformation matrix, and the influence of the defects that the real background is difficult to replace by the virtual construction background and the license plate distortion and the virtual construction background distortion are difficult to keep consistent on the enhanced image is avoided.

Description

technical field [0001] The present application belongs to the technical field of data processing, and in particular relates to a license plate image enhancement method, device, equipment and storage medium. Background technique [0002] For big data-oriented deep learning technology, in order to obtain high-precision algorithm models, deep learning networks need to use a large amount of labeled data for training. Supervised learning is the most important and widely used type of deep learning tasks. In supervised learning, the data sample set is the foundation, and the size and label quality of the data sample set directly determine the application effect of the algorithm model. [0003] In the current license plate recognition technology, in order to improve the accuracy of the algorithm model of license plate recognition, in addition to geometric transformation (such as flipping, rotating, cropping, deformation, scaling), color transformation (such as noise, blur, color tra...

AI Technical Summary

Problems solved by technology

[0003] In the current license plate recognition technology, in order to improve the accuracy of the algorithm model of license plate recognition, in addition to geometric transformation (such as flipping, rotating, cropping, deformation, scaling), color transformation (such as noise, blur, color transformation, erasure, filling) , sample overlay (such as pixel overlay, clipping overlay) and other traditional data enhancement methods, it is also proposed to achieve data enhancement through image synthesis to solve the difficulties in sample collection and workload in traditional data enhancement methods. Defects such as time-consuming

However, the inventors found that the existing data enhancement methods through image synthesis have defects such as perspective distortion, difficulty in replacing the real background with a virtual construction background, and difficulty in keeping the distortion of the artificial license plate consistent with the distortion of the virtual construction background. The gap between the enhanced data and the actual application scenario is too large, resulting in the problem that the algorithm model training accuracy is high but the actual application effect is poor

Images